Geoscience Reference
In-Depth Information
numObjects
In our example the method identii ed 192 grains, which is signii cantly less
than the 236 grains counted manually, reduced by the three objects that
overlap the borders of the image. Visual inspection of the color-coded image
generated by
bwboundaries
reveals the reason for the underestimated number
of grains. Two large grains in the middle of the image have been observed as
being connected, giving a single, very large grain in the i nal result. Reducing
the disk size with
strel
from
disk=1
to
disk=5
can help separate connected
grains. Larger disks, on the other hand, reduce the number of grains because
smaller grains are lost by i ltering. We now determine the areas of each of
the grains.
graindata = regionprops(labeled,'basic');
grainareas= [graindata(:).Area];
objectareas = 3^2 * grainareas * 367^(-2);
We then i nd the maximum, minimum and mean areas for all grains in the
image, in cm
2
.
max_area = max(objectareas)
min_area = min(objectareas)
mean_area = mean(objectareas)
h e connected grain in the middle of the image has a size of 0.16 cm
2
, which
represents the maximum size of all grains in the image. Finally, we plot the
histogram of all the grain areas.
clf
e = 0 : 0.0005 : 0.15;
histogram(objectareas,e)
xlabel('Grain Size in Millimeters^2')
ylabel('Number of Grains')
axis([0 0.1 0 30])
Several methods exist that partly overcome the artifact of connected
grains in grain size analyses. h e most popular technique for region-based
segmentation is the watershed segmentation algorithm. Watersheds in
geomorphology are ridges that divide areas contributing to the hydrological
budget of adjacent catchments (see Section 7.10). Watershed segmentation
applies to grayscale images the same methods used to separate catchments
in digital elevation models. In this application, the grayscale values are
interpreted as elevations in a digital elevation model, where the watershed
then separates the two objects of interest.
h e criterion commonly used to identify pixels that belong to one object
is the nearest-neighbor distance. We use the distance transform performed
